Goldman Sachs and Apollo debate AI buildout financing and private credit prospects

The real AI trade is moving off the screen

Jim Zelter’s message is simple enough to fit in one sentence and big enough to reshape how Goldman Sachs clients think about AI: the money is moving from software headlines to the infrastructure that makes the whole thing possible. Goldman Sachs used its April 16 episode with the Apollo president to frame AI not as a neat technology theme, but as a long-duration capital story built on private credit, data centers, power access, and financing structures that can absorb enormous checks.

That matters inside Goldman because it changes where the durable fees, relationships, and influence will come from. If the next phase of AI is about who can finance the buildout, then the winning desks are not just the ones talking about model breakthroughs or chip demand, but the teams that can price risk on assets with 10-year lives, structure leases, and line up capital for everything from compute clusters to grid upgrades.

Why the capital numbers are getting so hard to ignore

Goldman Sachs Research has already put a hard number on how fast expectations are moving. Consensus estimates for 2026 AI hyperscaler capital spending rose to $527 billion, up from $465 billion at the start of the third-quarter earnings season. That is not a small revision, and it tells you the market is still in the phase where each quarter seems to reset what “normal” AI spending looks like.

At the same time, Goldman has noted that the average stock-price correlation across large public AI hyperscalers has fallen from 80% to 20% since June, which suggests investors are starting to separate the winners from the rest. That shift matters for clients who have been treating AI as one single trade. The market is becoming less about buying the whole basket and more about identifying who owns the bottlenecks, who finances them, and who captures the economics when the buildout gets more selective.

What actually has to be built

Apollo’s own materials make the physical scale of the opportunity harder to hand-wave away. The firm says AI is accelerating demand for power, connectivity, data centers, clean power, and grid modernization, and it argues that global data center infrastructure may require several trillion dollars of investment over the next decade. Apollo Institutional has said capex across just four hyperscalers, Alphabet, Amazon, Meta, and Microsoft, is expected to exceed $350 billion in 2025, while nearly $3 trillion will be needed to support AI infrastructure through 2028, with more than half of that, around $1.5 trillion, coming from external capital.

That is the crucial detail for Goldman employees and clients: this is not just an internal tech budget problem for a handful of hyperscalers. It is an external financing problem that pulls in lenders, private credit managers, infrastructure investors, power developers, and structured-finance teams. Once the story becomes power supply, fiber routes, data halls, cooling systems, and the lease economics around GPU stacks, the conversation stops sounding like a venture pitch and starts sounding like a capital stack.

Apollo is already showing how the money gets raised

Apollo is not talking about this buildout in the abstract. On Jan. 7, 2026, Apollo-managed funds and affiliates said they led a $3.5 billion capital solution for Valor Compute Infrastructure to support a $5.4 billion acquisition and lease of data center compute infrastructure for a subsidiary of xAI Corp., including NVIDIA GB200 GPUs. That deal is a clean example of how AI infrastructure finance is being assembled: purchase the hardware, package the economics into a lease, and bring in outside capital to make the transaction work at scale.

That structure is exactly why private credit looks so relevant. Traditional bank balance sheets can help, but the sheer size and duration of the funding need creates room for private lenders and hybrid capital providers to step in with bespoke terms. For Apollo, the attraction is obvious: this is the kind of asset-backed, income-producing, long-duration financing that private credit was built to chase.

Goldman is trying to keep the frame broader than chips

Goldman’s own AI coverage suggests it is not interested in letting the market reduce the trade to semiconductors alone. The firm has argued that the next phase of the AI trade may broaden beyond chipmakers to AI platform stocks and productivity beneficiaries, which is a reminder that the returns from AI adoption are likely to spread unevenly across software, services, and corporate workflows. Goldman has also been cautious before, warning in April 2025 that infrastructure investment should be judged against falling model costs and the risk of overbuilding, much like the telecom boom.

That caution is important because it keeps the story honest. If model costs keep falling faster than expected, the buildout could outrun demand in some places, leaving investors with underutilized assets and weaker returns. Goldman’s April 2 episode, “AI Exchanges: Power Problems?”, shows the firm has been pushing beyond the hype cycle and into the constraints that actually determine who makes money.

What this means inside Goldman Sachs

For analysts, associates, and VPs, the practical takeaway is that AI is becoming a cross-coverage story with longer legs than a typical product cycle. The opportunity set now spans infrastructure finance, private credit, power, and the lease structures that sit between the balance sheet and the machine. That creates work for people who can translate a flashy AI narrative into something a client can actually underwrite, finance, and defend through a full market cycle.

It also changes what counts as expertise. In a firm like Goldman, where prestige often follows the hardest transactions and the deepest client relationships, the people who understand AI buildout financing may end up closer to the most important mandates than the people who only track the headlines. Zelter’s appearance on Goldman’s platform is a reminder that the AI race is no longer just about who builds the smartest model. It is about who finances the warehouses, power systems, and credit structures that keep the models running.